RE: Degraded performance of Replication

See comments below ------------------------------------------ Alexey Sonkin Senior Database Administrator > -----Original Message----- > From: Link, David A. [mailto:DALink@west.com] > Sent: Wednesday, January 21, 2004 3:43 PM > To: Alexey Sonkin; informix-list@iiug.org > Subject: RE: Degraded performance of Replication > > Answers to your questions are below. I appreciate your help. > > > IBM/Informix is currently pointing us towards a memory performance patch. > We will retry test tonight. > > -----Original Message----- > From: Alexey Sonkin [mailto:alexeis@grandvirtual.com] > Sent: Wednesday, January 21, 2004 1:50 PM > To: Link, David A.; informix-list@iiug.org > Subject: RE: Degraded performance of Replication > > Sounds strange to me. > > In the past, I saw considerable performance gain > after switching from 7.31ucXX (32-bit) to 7.31fcXX (64-bit) > on HP (V-class) > > Did You make any OS performance measurements (CPU and disk load) > during Your experiments? Yes and we had no bottle necks. > > Did You notice any change any change in the load pattern? No. > > What is Your major bottleneck: disk or CPU? Neither seem to be. > > Did You see any OS swapping/paging ? No swapping or paging > > Did You make any Informix-specific monitoring (partition access > statistics, disk access statistics) during Your experiments? No stats > collected other than O/S and replication throughput and transaction > durations. > > In case new server chooses different query plans You should > observe a change in partition (table/index) access statistics > (especially bufreads). Yes we saw an increase of bufreads. This might indicate the real problem: new engine has chosen different query execution plan, e.g. based on different index. You can try to find out what objects (tables/indexes) were mostly accessed with a simple SQL: Select * from sysmaster:sysptprof order by bufreads desc Compare the results for two engines! > > Did You make 'update statistics' after switching to 9.40? Yes, used the > same process(dostats). > > How did You transform data from 9.30 to 9.40 (ontape, re-load, etc?) > Unload and reloaded. > > ------------------------------------------ > Alexey Sonkin > Senior Database Administrator > > -----Original Message----- > From: Link, David A. [mailto:DALink@west.com] > Sent: Wednesday, January 21, 2004 8:48 AM > To: informix-list@iiug.org > Subject: Degraded performance of Replication > > We have a replicated table on 2 IBM P680's running AIX 4.3.3 and Informix > 9.30.UC2. > In our tests we have pushed through as many as 1600 tx/sec, which is > great. > We are planning on upgrading to AIX 5.2 and Informix 9.40UF3. > We are only able to get 1300 tx/sec on the same class of servers. > I would have thought that with the OLTP improvements in 9.40 that we would > have gotten better performance. > > The only thing we have changed is that we narrowed the table. We are going > to retest with the original table but I thought I would see if anyone > would > have have any suggestions? > Just looking at the incoming OLTP transactions the newer version seems to > run much better. I am just wondering what we could be missing. > > Thanks for any suggestions in advance. > > David Link > sending to informix-list